After incorporating fourth-order force constants, the R-squared value is markedly less than 1.0, and corresponding solutions are sought.

[Ztao0210]

dear developer,

For cubic-phase CsPbI₃.I have successfully skipped the initial 2000 MD steps by modifying the tdep_parse_output.py script, and extracted the force constants using the command:

extract_forceconstants -rc2 100 -rc3 5 -rc4 3 -s 50 > fc2_fc3_fc4.log

Nevertheless, the calculation results remain unsatisfactory. The R-squared value is only around 0.014, and the resulting phonon dispersion still exhibits severe imaginary frequencies. However, existing literature confirms that the cubic-phase CsPbI₃ structure is thermodynamically stable at 300 K.

R^2 stddev energy (meV/atom)

                     input:                 5.6477037
              second order:  0.01416        8.3209152
               third order:  0.01465        9.2915303
              fourth order:  0.03633       13.5186824

I am writing to ask whether extending the MD simulation duration and increasing the number of sampling points could resolve this issue.

I am also attaching input files and output files below.

CsPbI.zip

Best regards,
Tao

[mjv500]

Hi Tao, you have massive forces awk '{for (ii=1;ii<=3;ii++) {normf=($ii*$ii)**0.5; if (maxf < normf){print maxf; maxf=normf}} }; END{print maxf}' infile.forces ... 1.73768 1.75225 1.77236 1.78681 1.79315 1.79464 1.80731 of up to 1 eV/Angstrom, so the structure goes crazy. Either your MLIP is bad or you are using some too short AIMD, or something, but I don't think this is TDEP's responsibility, and it can not fix it or fit it properly. What I can not check easily is whether the infile.positions dynamics makes any sense and is oscillating around some reference position: missing this is usually the reason R^2 is low. You may also be melting the structure. M

…

On Sat, Jun 27, 2026 at 1:14 PM TAO ***@***.***> wrote: *Ztao0210* created an issue (tdep-developers/tdep#180) <#180> dear developer, I have successfully skipped the initial 2000 MD steps by modifying the tdep_parse_output.py script, and extracted the force constants using the command: extract_forceconstants -rc2 100 -rc3 5 -rc4 3 -s 50 > fc2_fc3_fc4.log Nevertheless, the calculation results remain unsatisfactory. The R-squared value is only around 0.014, and the resulting phonon dispersion still exhibits severe imaginary frequencies. However, existing literature confirms that the cubic-phase CsPbI₃ structure is thermodynamically stable at 300 K. R^2 stddev energy (meV/atom) input: 5.6477037 second order: 0.01416 8.3209152 third order: 0.01465 9.2915303 fourth order: 0.03633 13.5186824 I am writing to ask whether extending the MD simulation duration and increasing the number of sampling points could resolve this issue. I am also attaching input files and output files below. CsPbI.zip <https://github.com/user-attachments/files/29407462/CsPbI.zip> Best regards, Tao — Reply to this email directly, view it on GitHub <#180?email_source=notifications&email_token=ABWOT37C42LUBAZH5BVY7U35B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABWOT33OXDC2IO4YSE5AUND5B6UBDAVCNFSNUABFKJSXA33TNF2G64TZHM3DKMBQGUZDMNRXHNEXG43VMU5TINZVHAYDONRZGI4KC5QC> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/ABWOT344ZXSMUGZLRY5QHZL5B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android <https://github.com/notifications/mobile/android/ABWOT325TN5GZHGCDLQ3BUD5B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society Chair, Steering Committee, European Theoretical Spectroscopy Facility www.etsf.eu Alumnus Fellow, Young Academy of Europe yacadeuro.org Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office University of Utrecht, 3584 CC Utrecht ITP Secretariat: +31 30 253 5928 E-mail: ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

[mjv500]

PS: claude has helpfully pointed out that your displacements are huge compared to the reference positions. There is probably (also?) something wrong with the correspondence between ssposcar and the positions files On Sat, Jun 27, 2026 at 5:25 PM matthieu verstraete < ***@***.***> wrote:

…

Hi Tao, you have massive forces awk '{for (ii=1;ii<=3;ii++) {normf=($ii*$ii)**0.5; if (maxf < normf){print maxf; maxf=normf}} }; END{print maxf}' infile.forces ... 1.73768 1.75225 1.77236 1.78681 1.79315 1.79464 1.80731 of up to 1 eV/Angstrom, so the structure goes crazy. Either your MLIP is bad or you are using some too short AIMD, or something, but I don't think this is TDEP's responsibility, and it can not fix it or fit it properly. What I can not check easily is whether the infile.positions dynamics makes any sense and is oscillating around some reference position: missing this is usually the reason R^2 is low. You may also be melting the structure. M On Sat, Jun 27, 2026 at 1:14 PM TAO ***@***.***> wrote: > *Ztao0210* created an issue (tdep-developers/tdep#180) > <#180> > > dear developer, > > I have successfully skipped the initial 2000 MD steps by modifying the > tdep_parse_output.py script, and extracted the force constants using the > command: > > extract_forceconstants -rc2 100 -rc3 5 -rc4 3 -s 50 > fc2_fc3_fc4.log > > Nevertheless, the calculation results remain unsatisfactory. The > R-squared value is only around 0.014, and the resulting phonon dispersion > still exhibits severe imaginary frequencies. However, existing literature > confirms that the cubic-phase CsPbI₃ structure is thermodynamically stable > at 300 K. > > R^2 stddev energy (meV/atom) > > input: 5.6477037 > second order: 0.01416 8.3209152 > third order: 0.01465 9.2915303 > fourth order: 0.03633 13.5186824 > > I am writing to ask whether extending the MD simulation duration and > increasing the number of sampling points could resolve this issue. > > I am also attaching input files and output files below. > > CsPbI.zip <https://github.com/user-attachments/files/29407462/CsPbI.zip> > > Best regards, > Tao > > — > Reply to this email directly, view it on GitHub > <#180?email_source=notifications&email_token=ABWOT37C42LUBAZH5BVY7U35B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ABWOT33OXDC2IO4YSE5AUND5B6UBDAVCNFSNUABFKJSXA33TNF2G64TZHM3DKMBQGUZDMNRXHNEXG43VMU5TINZVHAYDONRZGI4KC5QC> > . > Triage notifications, keep track of coding agent tasks and review pull > requests on the go with GitHub Mobile for iOS > <https://github.com/notifications/mobile/ios/ABWOT344ZXSMUGZLRY5QHZL5B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> > and Android > <https://github.com/notifications/mobile/android/ABWOT325TN5GZHGCDLQ3BUD5B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. > Download it today! > You are receiving this because you are subscribed to this thread.Message > ID: ***@***.***> > -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society Chair, Steering Committee, European Theoretical Spectroscopy Facility www.etsf.eu Alumnus Fellow, Young Academy of Europe yacadeuro.org Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office University of Utrecht, 3584 CC Utrecht ITP Secretariat: +31 30 253 5928 E-mail: ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society Chair, Steering Committee, European Theoretical Spectroscopy Facility www.etsf.eu Alumnus Fellow, Young Academy of Europe yacadeuro.org Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office University of Utrecht, 3584 CC Utrecht ITP Secretariat: +31 30 253 5928 E-mail: ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

[mjv500]

Hello again Tao, I have (asked Claude to help) made a TDEP visualizer app, to look at the trajectories. https://github.com/tdep-developers/tdep-view Your data on CsPbI is already public in the present github thread, but do you agree that I use it as example data in the app? Please try it out, there are several backends for visualization, and also a mode to check if the average structure is close to the reference one: this is your problem, e.g. the first atoms are at 0 0 0 in positions but 0.5 0.5 0.5 in ssposcar, which is all wrong. I have made an issue in tdep (fortran) that this check should be done automatically before extractforceconstants to flag your error. Matthieu On Sat, Jun 27, 2026 at 7:22 PM matthieu verstraete < ***@***.***> wrote:

…

PS: claude has helpfully pointed out that your displacements are huge compared to the reference positions. There is probably (also?) something wrong with the correspondence between ssposcar and the positions files On Sat, Jun 27, 2026 at 5:25 PM matthieu verstraete < ***@***.***> wrote: > Hi Tao, > > you have massive forces > > awk '{for (ii=1;ii<=3;ii++) {normf=($ii*$ii)**0.5; if (maxf < > normf){print maxf; maxf=normf}} }; END{print maxf}' infile.forces > > ... > > 1.73768 > > 1.75225 > > 1.77236 > > 1.78681 > > 1.79315 > > 1.79464 > > 1.80731 > > > of up to 1 eV/Angstrom, so the structure goes crazy. Either your MLIP is > bad or you are using some too short AIMD, or something, but I don't think > this is TDEP's responsibility, and it can not fix it or fit it properly. > What I can not check easily is whether the infile.positions dynamics makes > any sense and is oscillating around some reference position: missing this > is usually the reason R^2 is low. You may also be melting the structure. > > > M > > > > > On Sat, Jun 27, 2026 at 1:14 PM TAO ***@***.***> wrote: > >> *Ztao0210* created an issue (tdep-developers/tdep#180) >> <#180> >> >> dear developer, >> >> I have successfully skipped the initial 2000 MD steps by modifying the >> tdep_parse_output.py script, and extracted the force constants using the >> command: >> >> extract_forceconstants -rc2 100 -rc3 5 -rc4 3 -s 50 > fc2_fc3_fc4.log >> >> Nevertheless, the calculation results remain unsatisfactory. The >> R-squared value is only around 0.014, and the resulting phonon dispersion >> still exhibits severe imaginary frequencies. However, existing literature >> confirms that the cubic-phase CsPbI₃ structure is thermodynamically stable >> at 300 K. >> >> R^2 stddev energy (meV/atom) >> >> input: 5.6477037 >> second order: 0.01416 8.3209152 >> third order: 0.01465 9.2915303 >> fourth order: 0.03633 13.5186824 >> >> I am writing to ask whether extending the MD simulation duration and >> increasing the number of sampling points could resolve this issue. >> >> I am also attaching input files and output files below. >> >> CsPbI.zip <https://github.com/user-attachments/files/29407462/CsPbI.zip> >> >> Best regards, >> Tao >> >> — >> Reply to this email directly, view it on GitHub >> <#180?email_source=notifications&email_token=ABWOT37C42LUBAZH5BVY7U35B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, >> or unsubscribe >> <https://github.com/notifications/unsubscribe-auth/ABWOT33OXDC2IO4YSE5AUND5B6UBDAVCNFSNUABFKJSXA33TNF2G64TZHM3DKMBQGUZDMNRXHNEXG43VMU5TINZVHAYDONRZGI4KC5QC> >> . >> Triage notifications, keep track of coding agent tasks and review pull >> requests on the go with GitHub Mobile for iOS >> <https://github.com/notifications/mobile/ios/ABWOT344ZXSMUGZLRY5QHZL5B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> >> and Android >> <https://github.com/notifications/mobile/android/ABWOT325TN5GZHGCDLQ3BUD5B6UBDA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBXGU4DANZWHEZDRJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. >> Download it today! >> You are receiving this because you are subscribed to this thread.Message >> ID: ***@***.***> >> > > > -- > > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > Professor Matthieu J Verstraete > Fellow, American Physical Society > Chair, Steering Committee, European Theoretical Spectroscopy Facility > www.etsf.eu > Alumnus Fellow, Young Academy of Europe yacadeuro.org > > Institute for Theoretical Physics (ITP) Department of Physics > Buys Ballot Gebouw/Building, Princetonplein 5, office > University of Utrecht, 3584 CC Utrecht > ITP Secretariat: +31 30 253 5928 E-mail: ***@***.*** > > Group web page (Liège): http://www.nanomat.ulg.ac.be/ > > Nanomat lab, Q-Mat center, Université de Liège > Département de Physique, Bat. B5a, 4/50 > Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium > Phone : +32 4 366 90 17 > > European Theoretical Spectroscopy Facility (ETSF) > > Mail : ***@***.*** > ***@***.*** > ***@***.*** > -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society Chair, Steering Committee, European Theoretical Spectroscopy Facility www.etsf.eu Alumnus Fellow, Young Academy of Europe yacadeuro.org Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office University of Utrecht, 3584 CC Utrecht ITP Secretariat: +31 30 253 5928 E-mail: ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society Chair, Steering Committee, European Theoretical Spectroscopy Facility www.etsf.eu Alumnus Fellow, Young Academy of Europe yacadeuro.org Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office University of Utrecht, 3584 CC Utrecht ITP Secretariat: +31 30 253 5928 E-mail: ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

[Ztao0210]

Dear Professor Matthieu,

Thank you very much for your attention and help regarding the issue I encountered with cubic CsPbI₃. I will try the newly developed tdep-view visualizer application as soon as possible.

I fully agree to make my CsPbI₃ dataset available as example data for this program. Please feel free to let me know if you require any additional data or supplementary files.

Best regards,
Tao